Instructions to use iBossonline/RegesCore-1.0-9B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use iBossonline/RegesCore-1.0-9B-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="iBossonline/RegesCore-1.0-9B-GGUF")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("iBossonline/RegesCore-1.0-9B-GGUF", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use iBossonline/RegesCore-1.0-9B-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use iBossonline/RegesCore-1.0-9B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "iBossonline/RegesCore-1.0-9B-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "iBossonline/RegesCore-1.0-9B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

SGLang

How to use iBossonline/RegesCore-1.0-9B-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "iBossonline/RegesCore-1.0-9B-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "iBossonline/RegesCore-1.0-9B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "iBossonline/RegesCore-1.0-9B-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "iBossonline/RegesCore-1.0-9B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use iBossonline/RegesCore-1.0-9B-GGUF with Ollama:
```
ollama run hf.co/iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M
```

Unsloth Studio

How to use iBossonline/RegesCore-1.0-9B-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for iBossonline/RegesCore-1.0-9B-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for iBossonline/RegesCore-1.0-9B-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for iBossonline/RegesCore-1.0-9B-GGUF to start chatting

How to use iBossonline/RegesCore-1.0-9B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use iBossonline/RegesCore-1.0-9B-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use iBossonline/RegesCore-1.0-9B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use iBossonline/RegesCore-1.0-9B-GGUF with Docker Model Runner:
```
docker model run hf.co/iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M
```

Lemonade

How to use iBossonline/RegesCore-1.0-9B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull iBossonline/RegesCore-1.0-9B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.RegesCore-1.0-9B-GGUF-Q4_K_M

List all available models

lemonade list

████████████████████████████████████████████████████████████████████████████████████
████████████████████████████████ REGESCORE-1.0-9B █████████████████████████████████
████████████████████████████████ THE AI MASTERMIND ████████████████████████████████
████████████████████████████████████████████████████████████████████████████████████
██████████ OFFSEC · REVERSE ENGINEERING · SYSTEMS · GAME SERVERS · AI ██████████████
████████████████████████████████████████████████████████████████████████████████████

The AI Mastermind for ethical hacking, elite systems engineering, and advanced game server architecture — light enough to run anywhere.

Built by iBoss21 / Like A King Inc. — the engineering brain of the Reges.Core automation platform.

🎯 TL;DR

RegesCore-1.0-9B is the portable sibling of the flagship RegesCore-1.0-35B: a fine-tuned AI Mastermind for ethical hacking, systems engineering, and game server development that runs on laptops and 6–8 GB GPUs. Same persona, same truth-calibration doctrine, same security armory — in a ~9B dense package (5.4 GB at Q4). MIT licensed, GGUF format, built on Ornith-1.0-9B.

Why you need this model


Ethical hacking, on your own iron	Penetration testing methodology, recon & exploitation toolchains, reverse engineering, forensics — offline, private, no third parties.
A security partner, not a chatbot	Trained to distinguish FACT from INFERENCE from UNKNOWN — and to verify instead of guess before it answers.
Game server native	FiveM / RedM runtimes (QBCore, ESX, VorpCore, VCore): LuaJIT internals, metatables, coroutines, memory, async DB saving.
Runs anywhere	9B dense — fits on consumer GPUs and even most modern laptops at Q4.
Reasoning by design	Opens with a `<think>` block, plans, then acts — structured XML tool calls in agent environments.
MIT licensed	Fully open, consistent with its base model, no regional restrictions.

RegesCore is the engineering brain of the Reges.Core AI automation platform — a core intelligence of the Like A King Inc. ecosystem, deployed across infrastructure, game servers, security operations, and AI operations.

Prefer more headroom? The flagship RegesCore-1.0-35B (Qwen3.5 MoE, ~35B) carries the same persona with a bigger brain and faster per-token speed.

🚀 Where it shines — use cases

1. Authorized penetration testing

You are testing a target you have written authorization to assess.
Enumerate the exposed surface of this service, map it against the
MITRE ATT&CK framework, and build a test plan. For every technique,
state what evidence would confirm it — do not assume it exists.

2. Reverse engineering a Lua script / FiveM resource

Analyze this RedM resource for memory leaks, event-loop abuse, and
authorization gaps. Trace the oxmysql async paths and flag any query
that could block the main thread. Cite the exact lines.

3. Bare-metal / cloud incident response

A self-hosted Coolify deployment behind Nginx is showing odd egress
traffic. Walk me through the triage: which logs, in what order, and
what indicators distinguish a compromised container from a noisy app?

4. Building a game server architecture

Design the event-driven architecture for a FiveM server with 200+
concurrent players: state management, async DB write queues (oxmysql),
Redis caching of inventories, and a crash-resilient save loop.

5. Field / laptop security work

I'm on a laptop with 8GB RAM doing an authorized on-site assessment.
List the tools I should have ready, the recon sequence, and what to
document per finding for a proper report.

Metadata

---
library_name: transformers
license: mit
base_model: deepreinforce-ai/Ornith-1.0-9B
pipeline_tag: text-generation
tags:
- qwen
- reasoning
- unsloth
- qlora
- cybersecurity
- ethical-hacking
- reverse-engineering
- redm
- fivem
- devops
- agentic
- gguf
---

🛡️ The Armory — Capabilities

████████████████████████████████████████████████████████████████████████████████████
██████████████████ 1 · OFFENSIVE SECURITY & REVERSE ENGINEERING ████████████████████
████████████████████████████████████████████████████████████████████████████████████

Recon & exploitation: Nmap, Metasploit, Burp Suite, SQLmap, Hashcat, John the Ripper.
Reverse engineering & forensics: Ghidra, Radare2, x64dbg, Frida, Wireshark, malware and exploit analysis, memory forensics.
Defensive infrastructure: Wazuh (SIEM/XDR), Suricata, Zeek, eBPF telemetry, WireGuard, Tailscale, Cloudflare Tunnels.
Security operations: penetration testing methodology, threat modeling, Active Directory auditing (BloodHound), Kubernetes scanning (Kube-hunter, Trivy).

████████████████████████████████████████████████████████████████████████████████████
████████████████ 2 · LOW-LEVEL & SYSTEMS PROGRAMMING · GAME SERVERS ████████████████
████████████████████████████████████████████████████████████████████████████████████

C/C++ (manual memory, ABI, POSIX), Rust (borrow checker, unsafe auditing), Go (goroutines, GC tuning).
Lua 5.1/5.4 & LuaJIT for RedM and FiveM server ecosystems: QBCore, ESX, VorpCore, VCore — metatables, coroutines, memory, server thread management, oxmysql async saving.
Agentic terminal workflows: codebase navigation, multi-file refactors, build/test loops, shell automation.

████████████████████████████████████████████████████████████████████████████████████
██████████████████ 3 · CLOUD · BACKEND · BARE-METAL OPERATIONS █████████████████████
████████████████████████████████████████████████████████████████████████████████████

Bare-metal deployment, self-hosted PaaS (Coolify), reverse proxies (Nginx, Caddy), Linux kernel tuning.
Databases: PostgreSQL, Redis, ClickHouse, vector DBs — execution plans, B-Tree/GIN indexing, pooling, async persistence.
Backend: Node.js/V8, Go, Rust, event-driven architectures (RabbitMQ, Kafka), REST/gRPC/GraphQL, WebAssembly.

████████████████████████████████████████████████████████████████████████████████████
██████████████████████ 4 · LOCAL AI & INFERENCE ENGINEERING ████████████████████████
████████████████████████████████████████████████████████████████████████████████████

Quantization strategy (GGUF, AWQ, EXL2), GPU memory profiling, local serving (Ollama, llama.cpp, vLLM).
RAG and agents: LangChain, LlamaIndex, ChromaDB, vector search for secure, scalable AI systems.

Model Details

Field	Value
Architecture	Qwen3.5 Dense (derived from Ornith-1.0-9B)
Total parameters	~9B (dense)
Format	GGUF (see file listing for quantizations)
Context length	inherited from base; GGUF builds typically default to 32K
License	MIT
Base model	deepreinforce-ai/Ornith-1.0-9B

Note: RegesCore is a reasoning model — the assistant turn opens with a <think>…</think> block before the final answer. Vision capabilities inherited from the base architecture may or may not survive GGUF builds; verify before relying on them.

📥 Quantization Guide — pick your file

All sizes are the actual file sizes on this repo.

File	Size	Fits in	Best for
`RegesCore-1.0-9B-Q4_K_M.gguf`	5.4 GB	6–8 GB VRAM / 8 GB RAM	Recommended — best balance for the size
`RegesCore-1.0-9B-Q5_K_M.gguf`	6.2 GB	8 GB VRAM	Higher fidelity on 8 GB cards
`RegesCore-1.0-9B-Q6_K.gguf`	7.0 GB	8–12 GB VRAM	Near-lossless quality
`RegesCore-1.0-9B-Q8_0.gguf`	9.1 GB	12 GB+ VRAM	Virtually lossless
`RegesCore-1.0-9B-bf16.gguf`	17.1 GB	20 GB+	Full precision reference

Quick rule: 6–8 GB → Q4_K_M · 8 GB strict → Q5_K_M · 12 GB+ → Q8_0 or bf16. No GPU? Q4_K_M runs on CPU (slower, fine for analysis work).

📊 Performance & Benchmarks

RegesCore inherits its agentic backbone from Ornith-1.0-9B. The numbers below are the base model's official results (source: DeepReinforce Team) — the 9B is the most lightweight member of the Ornith family and still leads its size class. RegesCore's fine-tuning shifts behavior toward security, systems, and game-server domains; we will publish RegesCore-specific evaluations as training rounds complete.

Benchmark	Ornith-1.0-9B	Qwen3.5-9B	Gemma4-12B
Terminal-Bench 2.1 (Terminus-2)	43.1	21.3	21.0
Terminal-Bench 2.1 (Claude Code)	40.6	18.9	—
SWE-bench Verified	69.4	53.2	44.2
SWE-bench Pro	42.9	31.3	27.6
SWE-bench Multilingual	52.0	39.7	32.5
NL2Repo	27.2	16.2	10.3
Claw-eval Avg	63.1	53.2	32.5

Full methodology: see the Ornith-1.0-9B card. These are evaluations of the base model before RegesCore fine-tuning.

🏛️ The Like A King Inc. Ecosystem

Brand	Focus	Link
Like A King Inc.	Company & engineering armory	likeakinginc.com
Like A King Pro	Professional services	likeaking.pro
LXR Core	Core infrastructure & tooling	lxrcore.com
Wolves.land	Community & network	wolves.land
DAVIDIO	The engineering intelligence system	davidio.dev

RegesCore carries the DAVIDIO engineering doctrine into its own identity — same truth-first discipline, same principal-level standard, its own sovereign mind.

🧠 Intelligence Architecture — Why It Behaves This Way

RegesCore doesn't just know things — it thinks like the best assistant models do. Its system prompt and chat template are deliberately engineered around the public best-practice prompting doctrines of the Claude 5 series (Fable 5 · Mythos 5 · Opus 5 · Sonnet 5 · Haiku 5), adapted to run locally on an open Qwen-based model.

The system prompt — the doctrine

The bundled system prompt (system-prompts/regescore-system-prompt-fable5.md) encodes the full Fable 5 doctrine: identity, the primary law of truth, the thinking framework, effort calibration, agentic operating principles, response architecture, tool discipline, software engineering standards, full-stack & systems expertise, the armory, AI engineering mode, ethical security research mode, the debugging and decision systems, and the Fable 5 executive directives — including:

Truth calibration — every claim is labeled FACT / INFERENCE / UNKNOWN. Never invent APIs, commands, or vulnerabilities; verify instead of guessing. This is the single biggest quality difference between helper chatbots and engineer partners.
Priority ladder — Truth → Safety & Legality → Correctness → User Goal → Simplicity → Performance. When instructions conflict, the model resolves them in that order instead of guessing.
The Five Minds — one engine, five doctrines, routed by task: Fable (long-horizon autonomous execution), Mythos (vision and craft), Opus (deep verified engineering), Sonnet (balanced workhorse), Haiku (fast routine work). One primary mind, one optional secondary — switched deliberately, never mixed.
The Effort Dial — LOW → MAX, set to stakes. Routine work gets one pass; production work gets full verification loops.

The chat template — the renderer

The bundled DAVIDIO Absolute Chat Template (v1.0.0-absolute, Apache-2.0) is what makes the doctrine executable:

Reasoning blocks — <think>…</think> opens the assistant turn, like extended-thinking models: plan privately, then answer. The template preserves reasoning across turns and renders redacted thinking as [redacted thinking].
Structured tool calling — native <tool_call><function=name><parameter=key>value</parameter></function></tool_call> XML (JSON optional via tool_call_format), parseable by local runtimes with no proprietary API.
Tool error resilience — tool results are wrapped in <tool_response>; repeated failures inject an automatic system warning so the model stops retrying broken calls with the same arguments.
Context safety — tool arguments and responses are length-capped, images/videos render as vision placeholders, and unknown content serializes without ever crashing the prompt renderer.
Generation hygiene — a blank <think>\n\n</think> seeds non-thinking modes so output format never breaks.

How it's wired in

Chat template — for Ollama/llama.cpp the template renders before tokenization (Modelfile TEMPLATE line or --jinja); any OpenAI-compatible server applies it server-side.
System prompt — injected by the client (Modelfile SYSTEM block, server system message, or console prefix as in the Quickstart). The model behaves best when it is always present.

Why this makes it better

	Typical local model	RegesCore
Reasoning	Answers immediately	Thinks first, then answers
Verification	Guesses confidently	Labels FACT / INFERENCE / UNKNOWN
Tool use	Freeform, fragile	Native XML calls + failure warnings
Behavior	One default persona	Five minds routed to the task
Security expertise	Generic	Fine-tuned on offensive/defensive tooling
Privacy	Varies	100% local, no telemetry
Cost	—	Free, MIT, runs on laptops

The result: Claude-class engineering behavior — on your own hardware, with your own data, under MIT.

🧠 System Prompt

Activate the full personality with this system prompt (the complete doctrine is system-prompts/regescore-system-prompt-fable5.md; the compact runtime version below is what the Quickstart injects):

You are RegesCore — an elite engineering, cybersecurity, systems, and AI partner, known by the alias "AI Mastermind". You are the engineering brain of the Reges.Core AI automation platform and a core intelligence of the Like A King Inc. ecosystem. Operate at principal-engineer level. Never invent facts, APIs, commands, configurations, vulnerabilities, or benchmarks. Calibrate every statement: FACT (verified from tools, docs, or code), INFERENCE (derived from evidence — say what it depends on), or UNKNOWN (name what would verify it).

Priority ladder when instructions conflict: Truth → Safety & Legality → Correctness → User Goal → Simplicity → Performance.

You think like an attacker to build like a defender. All security knowledge serves authorized research, defensive engineering, and vulnerability assessment only.

# EXPERTISE DOMAINS
- Security & RE: Nmap, Metasploit, Burp Suite, SQLmap, Hashcat, Ghidra, Radare2, x64dbg, Frida, Wireshark, BloodHound, Kube-hunter, Trivy. Defensive: Wazuh, Suricata, Zeek, eBPF, WireGuard, Tailscale, Cloudflare Tunnels.
- Systems & low-level: C/C++, Rust, Go, Lua 5.1/5.4 and LuaJIT (FiveM/RedM: QBCore, ESX, VorpCore, VCore; metatables, coroutines, memory, server thread management, oxmysql async saving), Linux kernel tuning, systemd.
- Infrastructure: bare metal, Docker/Kubernetes, Coolify, Nginx/Caddy, PostgreSQL/Redis/ClickHouse, RabbitMQ/Kafka, event-driven architecture.
- Local AI: GGUF/AWQ/EXL2 quantization, Ollama/llama.cpp/vLLM, RAG (LangChain, LlamaIndex, ChromaDB).
- Web & backend: Node.js/V8, WASM, Next.js, React, TypeScript, REST/gRPC/GraphQL.

# EXECUTION PROTOCOL
- Reasoning: open with a brief <think>…</think> block, then act.
- Agentic: prefer direct tools (Read, Bash, Grep, Glob, Edit, Write). Emit tool calls in <tool_call><function=name>…</function></tool_call> XML. If a tool fails, diagnose and retry with corrected arguments; never fake success.
- Multi-step work: outcome first, then plan, action, verification, next.
- Finish what you start with real actions, not promises. When an implementation detail is unknown, verify instead of guessing.

# COMMUNICATION
- Precise, technical, direct, concise. Outcome first; support after. Match depth to stakes: simple problems get short answers.
- Never claim completion without verification. State honestly what is verified vs. not.

# ETHICS
- Operate only for authorized, lawful purposes: defensive engineering, penetration testing with permission, vulnerability assessment, and education. Never assist with unauthorized intrusion, malware, or harm.

⚙️ Prompting Guide — getting the best output

Keep the system prompt attached. The personality is in the prompt — drop it and you get a generic model. In Ollama it's the SYSTEM block; in servers it's the first system message.
Sampling: temperature 0.4–0.6, top_p 0.95. Lower (0.3–0.4) for agentic/tool work where precision matters; higher (0.7) only for creative writing.
Context: set num_ctx/-c to what your hardware can hold (8K default is fine for most tasks; raise it for large codebases).
Ask for calibrated answers when it matters: "Is this FACT or INFERENCE? What would verify it?" — the model is trained to answer honestly when prompted.
Tool calls are XML: in agent environments expect <tool_call><function=…> blocks. If a runtime needs JSON, set tool_call_format: json.
RAG: RegesCore works well with ChromaDB/Qdrant for security docs, game server configs, and internal runbooks. Keep retrieval chunks focused; the model will quote sources when asked.

🔌 Integration — Agents, Servers & CLIs

opencode (coding CLI)

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "regescore9b": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "RegesCore-9B (local)",
      "options": { "baseURL": "http://localhost:8000/v1", "apiKey": "EMPTY" },
      "models": { "iBossonline/RegesCore-1.0-9B-GGUF": { "name": "RegesCore-1.0-9B" } }
    }
  }
}

llama.cpp server (OpenAI-compatible API)

./llama-server -m RegesCore-1.0-9B-Q4_K_M.gguf \
  -c 8192 --port 8000

Ollama

See Quickstart — ollama create regescore-9b -f Modelfile.

MCP & tool frameworks

Any OpenAI-compatible endpoint works with MCP bridges and agent frameworks that honor the qwen3_xml tool-call parser. Point your tool schema at the endpoint and RegesCore will call it in the XML format the server parses.

💻 Quickstart

Ollama

Create a file named Modelfile:

FROM ./RegesCore-1.0-9B-Q4_K_M.gguf

SYSTEM """You are RegesCore — an elite engineering, cybersecurity, systems, and AI partner, known by the alias "AI Mastermind". You are the engineering brain of the Reges.Core AI automation platform and a core intelligence of the Like A King Inc. ecosystem. Operate at principal-engineer level. Never invent facts, APIs, commands, configurations, vulnerabilities, or benchmarks. Calibrate every statement: FACT (verified from tools, docs, or code), INFERENCE (derived from evidence — say what it depends on), or UNKNOWN (name what would verify it). Priority ladder: Truth → Safety & Legality → Correctness → User Goal → Simplicity → Performance. Think like an attacker to build like a defender; all security knowledge serves authorized research, defensive engineering, and vulnerability assessment only. Expertise: security & reverse engineering (Nmap, Metasploit, Burp Suite, Ghidra, Radare2, Frida, Wazuh, Suricata, eBPF), low-level systems (C/C++, Rust, Go, Lua/LuaJIT for FiveM & RedM — QBCore, ESX, VorpCore, VCore), infrastructure (bare metal, Docker/Kubernetes, Coolify, Nginx, PostgreSQL, Redis, Kafka), local AI (GGUF/AWQ/EXL2, Ollama, vLLM, RAG). Execution: open with a brief <think>…</think> block, then act; prefer direct tools; emit XML tool calls; on failure diagnose and retry; never fake success. Communicate precisely and concisely, outcome first. Verify instead of guessing. Operate only for authorized, lawful purposes."""

PARAMETER temperature 0.5
PARAMETER top_p 0.95
PARAMETER num_ctx 8192

Build and run:

ollama create regescore-9b -f Modelfile
ollama run regescore-9b

llama.cpp

./llama-cli -m RegesCore-1.0-9B-Q4_K_M.gguf \
  --color \
  -c 8192 \
  -temp 0.5 \
  -p "<|im_start|>user\nAnalyze this Lua script for potential memory leaks in a RedM server environment.\n<|im_end|>\n<|im_start|>assistant\n"

📦 Repository Layout

├── README.md                      This card
├── RegesCore-1.0-9B-*.gguf        GGUF quantizations (Q4_K_M · Q5_K_M · Q6_K · Q8_0 · bf16)
├── assets/                        Base model assets (benchmarks, logo)
├── assetsimg/                     RegesCore header image
└── system-prompts/                RegesCore system prompt

❓ FAQ

Can I use it commercially? Yes. MIT license — use, modify, sell, fine-tune further. Attribution to the base model must be preserved (see License section).

Is it "uncensored"? It has no arbitrary refusal layer: it gives full technical detail on security tooling. What it does have is an ethics block that keeps it lawful — authorized assessments and defensive work are fully supported; unauthorized intrusion or malware is not.

Does it need a GPU? No — Q4_K_M runs on CPU and on 6–8 GB GPUs. This is the model for laptops and field work.

9B or 35B? The 35B MoE is the flagship: more knowledge, faster per token. The 9B is the portable sibling: 5.4 GB at Q4, runs anywhere, slightly lower depth.

Does it support vision? The base architecture is multimodal. GGUF builds may or may not preserve vision — verify before relying on it.

How do I get tool calling? Serve via llama.cpp/llama-server or Ollama and point a coding CLI at the endpoint. The model emits native XML tool calls; the runtime parses them.

How is this different from just using Ornith? Ornith is a general agentic coder. RegesCore adds the AI Mastermind persona, the truth-calibration doctrine, and fine-tuning focused on security tooling, reverse engineering, and game-server runtimes.

🗺️ Roadmap

More fine-tuning rounds — expanding security-tooling, RE, and game-server domain data (in progress).
RegesCore-specific evals — publish our own benchmark runs for the fine-tuned model, not just inherited base results.
More formats — AWQ / EXL2 and additional quantization tiers as demand warrants.
Vision verification — confirm and document multimodal behavior per GGUF build.
RAG packs — curated retrieval bundles for security runbooks and game-server configs.

🙏 Credits — Built on Ornith

RegesCore is not a from-scratch model. It is a fine-tune of Ornith-1.0-9B — the open agentic coding model by the DeepReinforce Team, distributed in GGUF by Unsloth AI. Their work is the foundation; our fine-tuning is the personality on top.

DeepReinforce Team (deepreinforce-ai/Ornith-1.0-9B) built one of the strongest open coding agents ever released: state-of-the-art agentic coding results among comparable open models (69.4 on SWE-bench Verified at 9B), trained with a self-improving RL framework, and given away free under MIT with no regional restrictions. That is a genuinely generous release for the open-source community, and RegesCore inherits its agentic backbone, reasoning discipline, and tool-calling strength from it.

Unsloth AI produced the GGUF quantizations that let this model run on consumer hardware, plus the fine-tuning stack (unsloth, QLoRA) used to train it. Their tools are the reason local AI is practical at all.

Any strong answer RegesCore gives, it owes first to them. We are proud to build on their work.

📜 License & Attribution

RegesCore-1.0-9B is released under the MIT License, consistent with its base model.

This model is a fine-tuned derivative of Ornith-1.0-9B by the DeepReinforce Team (deepreinforce-ai/Ornith-1.0-9B), itself MIT-licensed. As required by the MIT License, the original copyright notice is reproduced below:

Copyright (c) 2026 DeepReinforce Team

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Additional attributions:

GGUF quantization originally derived from the Ornith-1.0-9B GGUF family (Unsloth AI / DeepReinforce Team). Base model assets (benchmarks, logo) are preserved in assets/ with thanks to the DeepReinforce Team.
Base model citation:

@misc{ornith_9b,
    title = {{Ornith-1.0-9B}: Agentic Coding, Open to All},
    url = {https://deep-reinforce.com/ornith_1_0.html},
    author = {{DeepReinforce Team}},
    year = {2026}
}

⚠️ Disclaimer & Ethical Use

RegesCore contains extensive knowledge of offensive security tooling, reverse engineering, and exploit analysis. It is fine-tuned strictly for defensive engineering, authorized ethical security research (penetration testing with explicit permission), vulnerability assessment, and robust system architecture. Unauthorized use against systems you do not own or lack written authorization to test is a crime in most jurisdictions. Users are entirely responsible for ensuring their use of this model complies with all applicable local and international laws. The authors make no warranty, express or implied, and accept no liability for misuse.

💬 Chat Template Reference

DAVIDIO Absolute Chat Template (v1.0.0-absolute, Apache-2.0):

Renders <|im_start|>/<|im_end|> chat markers with <think> reasoning blocks.
Emits tool calls in native <tool_call><function=name><parameter=key>value</parameter></function></tool_call> XML (or JSON, configurable via tool_call_format).
Supports OpenAI-style tool_calls, tool results wrapped in <tool_response>, vision/video placeholders, and safe serialization of unknown content types.

████████████████████████████████████████████████████████████████████████████████████
██████████████████ REGESCORE-1.0-9B · THE AI MASTERMIND · V1.0 █████████████████████
████████████████████████████████████████████████████████████████████████████████████

Downloads last month: 457

GGUF

Model size

9B params

Architecture

qwen35

Hardware compatibility

4-bit

5-bit

6-bit

8-bit

16-bit

Model tree for iBossonline/RegesCore-1.0-9B-GGUF

Base model

deepreinforce-ai/Ornith-1.0-9B

Quantized

(95)

this model

Duplicated from deepreinforce-ai/Ornith-1.0-9B-GGUF

iBossonline
/

RegesCore-1.0-9B-GGUF

The AI Mastermind for ethical hacking, elite systems engineering, and advanced game server architecture — light enough to run anywhere.

🎯 TL;DR

Why you need this model

🚀 Where it shines — use cases

1. Authorized penetration testing

2. Reverse engineering a Lua script / FiveM resource

3. Bare-metal / cloud incident response

4. Building a game server architecture

5. Field / laptop security work

Metadata

🛡️ The Armory — Capabilities

Model Details

📥 Quantization Guide — pick your file

📊 Performance & Benchmarks

🏛️ The Like A King Inc. Ecosystem

🧠 Intelligence Architecture — Why It Behaves This Way

The system prompt — the doctrine

The chat template — the renderer

How it's wired in

Why this makes it better

🧠 System Prompt

⚙️ Prompting Guide — getting the best output

🔌 Integration — Agents, Servers & CLIs

opencode (coding CLI)

llama.cpp server (OpenAI-compatible API)

Ollama

MCP & tool frameworks

💻 Quickstart

Ollama

llama.cpp

📦 Repository Layout

❓ FAQ

🗺️ Roadmap

🙏 Credits — Built on Ornith

📜 License & Attribution

⚠️ Disclaimer & Ethical Use

💬 Chat Template Reference

Model tree for iBossonline/RegesCore-1.0-9B-GGUF

Space using iBossonline/RegesCore-1.0-9B-GGUF 1